Título: Um Modelo de Rede Neural Baseado na Circuitaria da Retina para Detecção de Padrões Eletroencefalográficos
Autores: Rodrigues, Marco Aurélio Benedetti; Cunha, Iria Pedroso da; Andriani, Vania A.; Marino Neto, José; Azevedo, Fernando Mendes de
Resumo: Different structural and functional attributes of extrafoveal and foveal circuitry in the retina allows respectively for differential detection and acuity processing capacities. Accurate object analysis is achieved by maintaining a detected visual object image in the foveal circuits through centrally controlled eye movements. The present study describes the development of a neuronal model based in these attributes, for detection of electroencephalographic (EEG) patterns of human sleep. In this model, two distinct artificial neural networks (ANNs) were developed. A dynamic feedfoward ANN, modelling functional attributes of extrafoveal retina (DANN), introduces a preprocessing step aimed at preliminary detection of patterns similar to those presented at the training stage. If any or both of these trained patterns, DANN outputs are directed to a second ANN (FANN), that simulates foveal retina properties, including lateral inhibition and “on” / “off” portions of the bipolar and ganglionar receptive fields, that forms self maps. Such a structure allow for higher acuity (or discriminative power) and are employed in final recognition of multiple patterns. This model was tested in segments of human sleep EEG and was shown to effectively detect and discriminate between K complexes and sleep spindles, typical of sleep stage II.
Código DOI: 10.21528/CBRN2001-079
Artigo em pdf: 5cbrn_079.pdf
Arquivo BibTex: 5cbrn_079.bib